Background: Patients with early-onset colorectal cancer (CRC) or those with multiple tumours associated with hereditary non-polyposis colorectal cancer (HNPCC) raise suspicion of the presence of germline DNA mismatch repair (MMR) gene mutations.
Aim: To analyse the value of family history, microsatellite instability (MSI) analysis and MMR protein staining in the tumour to predict the presence of an MMR gene mutation in such patients.
Methods: In 281 patients diagnosed with CRC before the age of 50 years or with CRC and at least one additional HNPCC-associated cancer, germline mutation analysis in MLH1, MSH2 and MSH6 was carried out with denaturing gradient gel electrophoresis and multiplex ligation-dependent probe amplification. MSI analysis with five consensus markers and MMR protein staining for MLH1, MSH2 and MSH6 were carried out in the tumours.
Results: 25 pathogenic mutations (8 in MLH1, 9 in MSH2 and 8 in MSH6) were found. MSI analysis missed three and immunohistochemistry (IHC) missed two mutation carriers. Sensitivities of family history, MSI analysis and IHC for the presence of a mutation were 76%, 82% and 88%, specificities were 64%, 70% and 84%, and positive predictive values were 19%, 23% and 38%, respectively. Multivariate analysis showed the highest odds ratio for IHC (38.3, 95% confidence interval 9.0 to 184). Prevalence of pathogenic germline MMR gene mutations in patients with CRC before the age of 50 years was 6% and in those with ⩾2 HNPCC-associated tumours was 22%. In the second group, no mutation carriers were found among the 29 patients who were diagnosed with their first tumour after the age of 60 years.
Conclusion: Family history, MSI analysis and IHC are indicative parameters to select patients with CRC for MMR gene mutation analysis. The data show that IHC is the best single selection criterion.
- ACI, Amsterdam criteria I
- ACII, Amsterdam criteria II
- CRC, colorectal cancer
- HNPCC, hereditary non-polyposis colorectal cancer
- IHC, immunohistochemistry
- MMR, mismatch repair
- MSI, microsatellite instability
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
- ACI, Amsterdam criteria I
- ACII, Amsterdam criteria II
- CRC, colorectal cancer
- HNPCC, hereditary non-polyposis colorectal cancer
- IHC, immunohistochemistry
- MMR, mismatch repair
- MSI, microsatellite instability
Hereditary non-polyposis colorectal cancer (HNPCC) or Lynch syndrome is a genetic disorder caused by germline mutations in DNA mismatch repair (MMR) genes, in particular in MLH1, MSH2 and MSH6. HNPCC is characterised by the early onset of gastrointestinal and urogenital cancers, especially colorectal and endometrial cancer. Synchronous and metachronous cancers are common, and multiple colorectal cancers can be observed in 20–40% of people with HNPCC.1–3
Originally, HNPCC was a clinical diagnosis, based on the Amsterdam criteria I (ACI).4 In 1999, the ACI were superseded by the Amsterdam criteria II (ACII), which included extracolonic HNPCC-associated tumours.5 After the discovery of germline MMR gene mutations as the cause of HNPCC, it soon became clear that these criteria were too stringent, as a large proportion of families carrying germline MMR gene mutations did not fulfil either ACI or ACII. Consequently, several groups formulated new sets of criteria to select patients for mutation analysis in the MMR genes.6,7,8,9,10,11,12,13,14,15,16 The best known are the Bethesda criteria, published in 1997 and revised in 2004.17,18 Patients fulfilling one of these criteria should be selected for microsatellite instability (MSI) analysis of their tumours. In the case of MSI, a patient should be offered mutation analysis. Some of the (revised) Bethesda criteria are based on the family history, and others are based on the patient’s own history (eg, colorectal cancer (CRC) before the age of 50 years and multiple HNPCC-associated cancers, irrespective of age).18 As self-reported family history is often not accurate, it is important to recognise potential MMR gene mutation carriers independent of the family history.19,20
According to the Bethesda criteria and most of the other sets of criteria, mutation analysis should be restricted to those patients whose tumour shows a high degree of MSI. However, a proportion of cancers of MSH6 mutation carriers were recently shown not to have MSI.21,22 Thus, some MSH6 mutations will be missed using this strategy. Alternatively, MMR protein staining of the tumour could be a valuable and possibly a better method of selecting patients for mutation analysis. Mutation analysis should be offered to those patients whose tumours show absence of staining of one of the MMR proteins.10,14 An advantage of this method is that absence of staining for one of the MMR proteins directly points towards the putatively mutated gene.
In view of these considerations, we determined the value of MSI analysis and MMR protein staining in tumours for the prediction of the presence of a germline MMR gene mutation in a large group of patients fulfilling at least one of the above-mentioned Bethesda criteria, independent of family history—that is, colorectal cancer (CRC) before the age of 50 years or at least two HNPCC-associated cancers. The results were compared with the predictive values of family history. The results should enable formulation of rational guidelines regarding whether or not to offer mutation analysis to a patient with cancer who is at risk of HNPCC.
PATIENTS AND METHODS
Patients diagnosed with CRC before the age of 50 years and patients with ⩾2 HNPCC-associated cancers, including at least one CRC, irrespective of age and family history, were invited to participate in this study. On the basis of data available at the time of initiation of this study in 1996, colorectal cancer, endometrial cancer, cancer of the small bowel, stomach, pancreas, biliary tract and ovaries, and transitional cell cancer of the pelvis, ureter and bladder were considered to be HNPCC associated. Using the data of the regional cancer registry of the Comprehensive Cancer Center North-Netherlands from 1989 onwards, doctors at the participating hospitals and general practitioners invited patients to take part in the study and referred them to the study coordinator. Patients newly diagnosed from September 1997 onwards were offered participation in the same way, as well as patients who were diagnosed before 1989 with CRC before the age of 50 years or with CRC and at least one other HNPCC-associated cancer, who came to our attention for whatever medical reason. Inclusion ended by December 2000. Patients gave written informed consent after written and verbal pretest counselling. Blood (20 ml) was collected for DNA isolation. A thorough family history for HNPCC-associated cancers was recorded. Clinical data were reviewed. Cancer material was obtained and histology was revised. With permission of the patients involved, medical records of relatives with HNPCC-associated cancers were collected, whenever possible, to verify the nature of the reported cancers. The participating patients were informed about the results of the genetic test, if they so wished. In such instances, they received verbal post-test counselling and a written summary.
Another source of patients was the Department of Clinical Genetics, Groningen, The Netherlands, where patients were referred to and counselled if they were suspected to have HNPCC, from 1985 to December 2000. The referral region of this department is the same as the area covered by the above-mentioned cancer registry. With permission of the patients involved, information on family history and on the results of MSI analysis, immunohistochemical and mutation analysis were used for this study. The medical ethics committees of the University Medical Center Groningen, Groningen, The Netherlands, and other participating hospitals approved the study.
Mutation analysis of the MLH1, MSH2 and MSH6 genes was carried out on DNA isolated from peripheral blood lymphocytes by denaturing gradient gel electrophoresis, followed, in the case of aberrant band patterns, by direct sequencing of independently amplified polymerase chain reaction products as described previously.23 Validating homemade denaturing gradient gel electrophoresis systems for more than 40 different genes and testing hundreds of proved variants resulted in a sensitivity of the electrophoresis of almost 100% in our laboratory.24 For the detection of large deletions (exonic deletions or deletions of a complete gene) and duplications, we used the MLH1/MSH2 exon deletion multiplex ligation-dependent probe amplification test (MRC-Holland, Amsterdam, The Netherlands).25 These data, for cases that had deletions of ⩾1 exon in the MLH1 or MSH2 gene, were confirmed by Southern blot analysis.26
For the purpose of this study, pathogenic mutations were defined as changes in the gene sequences that cause allele inactivation either because of the production of a truncated protein product or because of an exonic deletion or duplication. Initially, mutation analysis for the three genes was carried out in every patient, but in the last one third of patients, mutations in MLH1 and MSH2 were sought for only those with an MSI-high tumour. As a result, in 60 of the 281 patients, mutation analysis was carried out only in MSH6, and not in MLH1 and MSH2.
Microsatellite instability analysis
MSI markers were used as previously defined, including two mononucleotide repeats (BAT25 and BAT26) and three dinucleotide repeats (D2S123, D5S346 and D17S250).27 DNA was extracted from formalin-fixed, paraffin-wax-embedded tumour sections. For MSI analysis, control DNA was obtained from normal tissue from paraffin-wax-embedded sections or from peripheral blood lymphocytes from the same patient. MSI analysis was carried out as described previously.21 Cancers were classified as “MSI high” when ⩾2 markers showed MSI and as “MSI low” when ⩽1 marker showed MSI. As a limited number of markers were analysed, the classification “microsatellite stable” was not used.
Immunohistochemistry for the MLH1, MSH2 and MSH6 proteins was carried out as described previously.21 Protein expression in normal tissue adjacent to the cancer served as an internal positive control. The sections were scored as either negative (ie, absence of detectable nuclear staining of cancer cells) or positive for MLH1, MSH2 and MSH6 staining. The staining was scored blinded to the MSI or mutation status.
Predictive values of different strategies to select patients for mutation analysis
The variables family history and results of MSI and immunohistochemisty (IHC) were analysed and different selection strategies were investigated to determine the optimal way to select patients for mutation analysis. The positive and negative predictive values, sensitivity and specificity of each strategy, and 95% confidence intervals (CIs) were calculated. In addition, we calculated the same values for “referral”; referral is defined as a patient having been referred to and counselled by a clinical geneticist versus inclusion within the framework of the study. For these calculations, the presence of a pathogenic germline mutation was the gold standard.
For statistical analysis, the χ2 test and the two-sided t test were carried out and p<0.05 was considered to be significant. The independence of the variables family history, MSI, IHC and referral for detecting MMR gene mutations was examined using multivariate analysis with backward selection.
A total of 281 apparently unrelated patients were included. Table 1 summarises their clinical data. Table 2 shows the characteristics of their first-degree relatives with cancer. Families of 21 index patients fulfilled the revised ACII.5 However, not all cancers were histologically verified in 15 of these families.
A total of 18 different pathogenic mutations were identified in 25 patients (8 in MLH1, 9 in MSH2 and 8 in MSH6; 25/281, 8.9%; table 3). All mutations were found in the 252 patients who had their first cancer diagnosed before the age of 60 years. Of the 17 MLH1 and MSH2 mutations, 15 occurred in patients who had their first cancer diagnosed before the age of 50 years, in contrast with four of the eight MSH6 mutations. The mean age of the mutation carriers at the time of diagnosis of the first cancer was 42 (range 13–56) years.
Of the 25 mutation carriers, 19 had a first-degree relative with an HNPCC-associated cancer. Thus, 19 of 111 (17%) patients with a positive first-degree family history proved to have a mutation, versus 6 of 170 (3.5%) of the others (p<0.001). Only 9 of the 25 mutation carriers came from the 21 ACII-positive families. Thus, the sensitivity of the revised Amsterdam criteria in this study was 36% and the specificity was 95%. Among those referred for genetic counselling at the Department of Clinical Genetics, a larger proportion of patients proved to be mutation carriers than among those who were directly referred to the study (11/69, 16% v 14/212, 6.6%; p = 0.018). The mean age at diagnosis of CRC in relatives of mutation carriers was 47.4 years, whereas it was 63.2 years in relatives of index patients without a mutation (p<0.001). The mean age at diagnosis of endometrial cancer was not different between the two groups (49.2 v 54.5 years; p = 0.138).
Nineteen missense mutations of unknown pathogenicity (7 in MLH1, 4 in MSH2 and 8 in MSH6) were detected in 18 patients. These missense mutations were not present in 200 healthy Dutch controls. One patient had an MSH6 missense mutation combined with a pathogenic MLH1 mutation, and another an MLH1 missense combined with an MSH6 missense mutation. For the missense mutation carriers, the mean age at onset of the first cancer was 45 (range 19–65) years. Seven of the 18 missense mutation carriers had a positive first-degree family history. As we do not know whether these missense mutations contribute to the disease phenotype, we excluded them from the study calculations. The clinical data on the missense mutation carriers are summarised in table A provided online at http://gutjnl.com/supplemental.
Microsatellite instability analysis
Tumour material was available for MSI analysis for 260 of the 281 patients. In 222 patients, only one cancer was available, whereas two cancers were available in 37 and three in one patient. Ninety seven patients had at least one MSI-high tumour. Among these patients, a higher frequency of mutation carriers was found than among the 163 patients with only MSI-low cancers (22/97 v 3/163, p<0.001). Three MSH6 mutation carriers, remarkably all with the 649–650insT mutation, had only MSI-low cancers.
When more than one cancer per patient was analysed for MSI, a discrepancy between the MSI results occurred in 9 of 38 cases, including three mutation carriers (patients 1, 9 and 25; table 3).
Interpretable tumour specimens for MLH1 and MSH2 immunostaining were available from 209 patients, whereas those for MSH6 immunostaining were available from 183 patients. Absence of at least one of the MMR proteins was seen in tumours from 50 patients. Complete IHC results were available for 17 of the 25 mutation carriers (only these cases are included in calculation of the predictive values). In all 20 cases with staining absent for one of the MMR proteins, absence of the MMR proteins was concordant with the corresponding mutation. Two mutation carriers (MLH1, case 6; and MSH6, case 24) were missed when IHC would have been used as a selection criterion. In 4 of the 24 patients from whom multiple cancers could be analysed, discrepancy in the IHC results was found, including two MSH6 mutation carriers (patients 18 and 23).
Predictive values of family history, MSI, IHC and referral for presence of a mutation
Table 4A shows the sensitivity, specificity and positive and negative predictive values of the different variables to predict a mutation. Calculations are based on data from the 174 patients from whom a complete set of variables was available. Similar calculations were made for MSI and IHC after preselection based on family history (table 4B) and for IHC after preselection based on MSI analysis (table 4C). The predictive values were also calculated for the three protein stainings separately (table 4D).
In the multivariate analysis with backward selection, a positive family history and negative staining of an MMR protein were found to be independent variables. The variables “referral” and MSI-high phenotype are not independent determinants. Absent staining of an MMR protein showed the highest odds ratio (table 4E).
This study was conducted primarily to define parameters that can reliably predict the presence of MMR gene mutations in two specific groups of patients with CRC suspected of having HNPCC—that is, patients with CRC before the age of 50 years and those with a CRC and at least one other HNPCC-associated cancer, irrespective of age. The main conclusions that can be drawn from this study are as follows. (1) Immunohistochemical staining for the MMR proteins MLH1, MSH2 and MSH6 is the best single method to select patients with CRC, suspected of having an MMR gene mutation, for mutation analysis. (2) The low prevalence of MMR gene mutations in patients with CRC below the age of 50 years justifies preselection by IHC and/or MSI analysis before mutation analysis is carried out. (3) The occurrence of ⩾2 HNPCC-associated cancers in patients aged ⩾60 years is rarely due to germline MMR gene mutations.
Recently, five large studies were published that also focused on the relative values of MSI analysis and IHC to detect potential germline MMR gene mutation carriers.7–9,11,28 The conclusion from all five studies and from several previous studies was similar to that from ours—namely, IHC is at least equally valuable for this goal as MSI analysis. There are, however, several differences between our study and the other five.
In three of the studies, immunostaining for the MSH6 protein and mutation analysis for germline mutations in MSH6 were not carried out.8,9,11 Although less prevalent in typical AC-positive families, our results suggest that MSH6 mutations occur at about the same rate as those in MLH1 and MSH2 in a less selected patient population. In the study by Hampel et al,8 mutation analysis for MSH6 was carried out only in patients whose tumours showed MSI (low or high) or who were considered to be at high risk for HNPCC on the basis of clinical criteria and whose tumours lacked MMR gene expression. Although, probably, most mutations were detected by this approach, some MSH6 mutations could have been missed. Only Southey et al7 carried out IHC for MSH6 in all tumours.
Stormorken et al28 included only patients who were referred for genetic counselling. This may have influenced the resulting predictive values. The same is true for the sensitivity and the positive predictive value of the Amsterdam criteria, which can be expected to be high in a selected population like the one studied by Stormorken et al. We observed that only 9 of 25 mutation carrier families fulfilled the revised Amsterdam criteria. Also, Hampel et al8 found only a small minority (3/23) of the mutation carriers’ families fulfilling the Amsterdam criteria.
A limitation of many studies on this topic is that selection criteria are compared with each other, without establishing the gold standard—that is, the presence of mutations—in all participants. This is also true for four of the five studies mentioned above. Although the assumption that all pathogenic mutations of MLH1 and MSH2 lead to MSI can be well defended, a proper calculation of the sensitivity, specificity and predictive values requires a true gold standard. We carried out mutation analysis for all three genes in the first two thirds of included patients and did not find any clearly pathogenic mutation in MLH1 or MSH2 in patients with MSI-low tumours.
The value of IHC depends partially on the quality of the nuclear stainings and the experience of the pathologist.12 The results are sometimes not interpretable because of absence or low intensity of the nuclear staining in tumour and normal cells.29 Mangold et al30 described the phenomenon of weak nuclear MLH1 staining in tumours of patients with MLH1 mutations, which can be observed as false-positive staining. Experienced pathologists should be able to recognise these tumours with weakly positive MLH1 staining as tumours from possible MLH1 mutation carriers. On the other hand, MSI analysis gives difficulties as well. The quality of tumour DNA isolated from paraffin-wax-embedded tumours may be poor and not usable for MSI analysis. Although seldom mentioned, interobserver variation also occurs when scoring MSI. About 15% of sporadic colorectal cancers show MSI and MLH1-negative stainings as a result of hypermethylation of the promoter region of MLH1.31 Table 4D shows that MLH1 staining has a low positive predictive value. The consequence of this is that more patients will be screened for MLH1 mutations with a negative outcome than is the case for the other two genes.
Apart from IHC and MSI analysis, we also evaluated the first-degree family history for HNPCC-associated cancers as an indicator of an MMR gene mutation. As expected, most mutation carriers had a first-degree family member with such a tumour. Nevertheless, 6 of the 25 (almost 25%) mutation carriers had a negative first-degree family history. This indicates that family history alone is an insufficient indicator of being a mutation carrier and we should not rely on it alone when deciding whether or not to carry out mutation analysis. We further calculated the predictive values for combinations of family history, MSI analysis and IHC. Although the specificity and positive predictive value both increased from this, the sensitivity decreased markedly. Therefore, when a patient fulfils one of the criteria used in this study, IHC or MSI analysis should be carried out without further selection. Both MSI and IHC analysis are proper selection tools, but when the family history raises high suspicion for HNPCC, both methods should be used to avoid missing mutation carriers.
Terdiman et al32 noticed that referral for genetic counselling because of the suspicion of hereditary cancer was an independent risk factor for the presence of an MMR gene mutation when compared with population-based patients with CRC. We could not confirm that finding, although we also found a higher prevalence of mutations in patients referred to the clinical geneticist. A positive family history in most of those referred to a clinical geneticist is probably the determinant of this finding.
Our group of 224 patients with CRC at <50 years of age is the largest group so far reported in a single study. Most of these patients were referred specifically to participate in this study, whereas a minority had been referred to a clinical geneticist because of a perceived high risk for hereditary cancer. All patients came from the same geographical area and had their tumours diagnosed in approximately the same period. We therefore think that this group can be considered representative of all patients with CRC before the age of 50 years. Pathogenic mutations were detected in 14 of the 224 (6%) patients. MLH1 and MSH2 mutations were identified in 11 (5%) patients. This is similar to the findings of Pinol et al11 and in accordance with prevalences of 0–12% in other small, otherwise unselected groups of such patients.33–38 Owing to the low mutation frequency, it is, in our view, justified to carry out only mutation analysis in such patients if a screening by MSI analysis or IHC analysis suggests a mutation.
This may be different for people with ⩾2 HNPCC-associated cancers, including at least one CRC. In this group of 79 people, we did not find any mutation in the 29 patients who had their first tumour diagnosed after age 60 years (19 of these had only CRCs and 10 had CRC and at least one other HNPCC-associated cancer). It is therefore questionable whether such patients should be screened for HNPCC and thus should be included in the Bethesda criteria. However, our group of patients was still rather small and further studies should be conducted before an age criterion for these patients can be confidently introduced. Nevertheless, when this subgroup is excluded from consideration, mutations were found in 17 of 50 patients with ⩾2 cancers, of which at least one was diagnosed before age 60 years. Such a high occurrence may justify direct mutation analysis, without a prescreen for MSI or IHC.
Recently it was reported that the mean age at diagnosis of CRC of mutation-positive relatives of probands, identified in a population-based study of patients with CRC, was 61.2 years, much higher than that reported so far for mutation-positive people, which is about 45 years.39 It is striking though, that the mean age at diagnosis of the probands in Hampel et al’s39 study was 44 years. As patients in our study were collected irrespective of their family history, a similarly higher age at diagnosis of CRC in the relatives of our patients might have been expected. This was not the case as the mean age was 47.4 years. The fact that we did not find any mutation carriers in patients with ⩾2 cancers, all diagnosed above age 60 years, also does not support the findings of Hampel et al39. Hence, their findings need confirmation in other populations, before any change in counselling practices is considered.
Only 8 of 25 supposedly mutation-carrying parents had CRC at the time of the study. Most of these 25 parents were older (⩾70 years) at that time and it was not expected that many of them would still develop CRC. This observation is in accordance with recent new calculations on the CRC risk in HNPCC—that is, 26.7% and 22.4% for men and women, respectively.40
A limitation of our study is the number of patients for whom a complete set of variables required for the calculations was available (n = 174). Although smaller than the populations in the studies by Engel et al9 (n = 1119) and Pinol et al11 (n = 287), we think that the number of patients in our study was large enough for reliable calculations. Another limitation might be the fact that we have missed mutations, as we included only the major HNPCC-related genes in our genetic screening. Mutations in other MMR genes, such as PMS2, cannot be ruled out.41–45 Furthermore, we did not screen for mutations in the regulatory sequences (such as the promoter regions) of MLH1, MSH2 and MSH6, nor did we look for genomic rearrangements in MSH6.46,47 However, as such mutations probably represent only a small proportion of disease-causing MMR gene mutations, this will not markedly alter the estimated prevalence of germline MMR gene mutations among patients with CRC. Finally, we cannot exclude survival bias in our study. Patients with MSI-high tumours have a better prognosis than those with MSI-low tumours, and, as we included some of the patients retrospectively and inclusion for the study required a blood sample for mutation analysis, the real mutation frequency might be somewhat lower.
The equal or even better potency of IHC to recognise patients with an MMR gene mutation does not apply to the detection of missense variants of currently unknown pathogenicity. In our study, 5 of 18 patients with such a variant had an MSI-high tumour, whereas IHC showed normal staining of the protein, corresponding with the variant gene, in all 14 cases where IHC was available (table A provided online). For some of the known missense variants, the evidence for pathogenicity is strong, but for most of them this is questionable. Functional assays of these variants may elucidate their pathogenic potential, but these have not yet been developed to a stage of routine application.
In conclusion, this study confirms that IHC staining for the MMR proteins is the best single method to select patients with CRC, suspected of having an MMR gene mutation, for mutation analysis. Such selection is mandatory for most patients with CRC before the age of 50 years, in view of the low prevalence of mutations in those patients. The value of genetic analysis in patients with multiple HNPCC-associated cancers, all occurring above age 60 years, is questionable.
We thank, in addition to the referring specialists in our hospital, the following specialists for patient referral: MA Alleman, JJ van Beek, JHW Bergmann, SB Bong, RWFR van den Broek, JG Bruyel, JW van Capelle, JLLM Coenen, GMJ DeKoninck, PFJ Donderwinkel, HH de Haan, LD de Haan, M van Haastert, H van der Heide, C Hoff, J Jager, HJA Jebbink, SS Klop, JJ Kolkman, WG Kootstra (deceased), L Meerman, HCJ van der Mijle, GF Nelis, M Oudkerk Pool, EGJM Pierik, W van Rooyen, CGBM Rupert, P Spoelstra, W Tel, JC Thijs, NP Tjon Pian Gi, JH Tomee, J Vecht, RJ Vermeer, RJ Vierhout, RA Weemhoff, EEM Weernink, EJGM Wichgers, HJ Wolters, LFA Wymenga and LS van der Zouwen. We thank J Schakenraad and M Schaapveld, Comprehensive Cancer Centre North-Netherlands, Groningen, The Netherlands, for selection of patients fulfilling our inclusion criteria from the regional cancer registry. We also thank RGJ Mensink and KA Kooi for their help with the mutation analysis, T van der Sluis and W Boersma for the immunohistochemical analysis and NJW Arts for her assistance in carrying out Southern blot analysis.
↵* RCN and MJWB contributed equally to this work.
Published Online First 24 April 2006
Funding: This work was supported by grant numbers RUG 1997–1544 and RUG 2002–2678 from the Dutch Cancer Society.
Competing interests: None.
Ethical approval: The medical ethics committees of the University Medical Center Groningen, Groningen, The Netherlands, and other participating hospitals approved this study (reference number MEC 97/02/037).